Nature's new rules: You can use ChatGPT to write papers, but not to be listed as an author

​This article is reprinted with the authorization of AI New Media Qubit (public account ID: QbitAI). Please contact the source for reprinting.

Facing ChatGPT, Nature finally couldn’t sit still.

This week, the fate of this authoritative academic publishing organization has been clarified regarding a series of issues such as ChatGPT’s ghostwriting of academic articles and being listed as an author.

Specifically, Nature lists two principles:

(1) Any large language model tool (such as ChatGPT) cannot become the author of the paper;
(2) If relevant tools have been used in the creation of the paper, the author should clearly state in the "Methods" or "Acknowledgments" or appropriate section.

The above requirements have now been added to the author submission guidelines.

#In recent times, ChatGPT has become more and more involved in the academic and research circles.

In a paper on the anti-aging application of rapamycin in December last year, ChatGPT was listed as a work, causing controversy in the industry. In addition to this article, there are many studies listing ChatGPT as the author.

Nature has also noticed this situation. A survey they conducted in December showed that 20% of the 293 professors interviewed had discovered or witnessed students using ChatGPT to complete assignments or papers. More people are worried about this.

This time, the statement issued by Nature is precisely hoping to characterize various disputes.

Multiple papers have listed ChatGPT as the author

ChatGPT was launched by OpenAI at the end of November last year because its performance significantly surpassed previous large languages. Model (LLM) quickly became the No. 1 "Sky Monkey" at the end of the year and the beginning of the year.

In the fields of new media, film and television, software development, game interaction and other fields, ChatGPT is quickly used to assist production and improve efficiency.

Xueyan Circle is no exception.

According to Nature statistics, there are at least 4 papers using ChatGPT and listing it as an author.

One of the preprints was published on the medical preprint database medRxiv in December 2022. This paper studies the performance of ChatGPT on the United States Medical Licensing Examination. Although the research is about ChatGPT, ChatGPT is also listed in the author column.

Another paper was published in the journal "Nurse Education Practice" on the pros and cons of open artificial intelligence platforms in nursing education. Likewise, ChatGPT is listed as an author:

The third paper comes from Insilico Medicine, an AI drug research and development company. It is about a new drug rapamycin and was published in "Oncoscience". Likewise, ChatGPT is listed as the author.

The fourth article is a bit "old" and was published in June 2022. The topic discussed is-the ability of AI to generate papers. The AI ​​listed in the author column is not ChatGPT, but GPT-3, which was released earlier.

Although the above research content is somewhat related to generative language models, for serious scientific research activities, use "research objects" to write papers and list them as authors Column, inevitably triggered controversy and doubts.

And even if AI is not listed as the author, the use of ChatGPT in academic and research circles is becoming more and more common.

As mentioned earlier, Alex Zhavoronkov, CEO of Insilico Medicine, an AI drug research and development company, revealed that his institution has published more than 80 papers generated by AI tools.

British professor Mike Sharples has been paying attention to the impact of generative AI on academic research. Not long ago, he personally demonstrated on Twitter how to use a large language model to generate an academic paper in 10 minutes. He described step by step how he used AI to generate a summary based on the title, which also triggered a lot of discussion.

It is more common for students to use ChatGPT to help write papers, generate code, and complete assignments. Currently, educational institutions in some areas of the United States have banned the use of this tool.

Just last week, a student at Northern Michigan University relied on ChatGPT to write an essay with the highest score in the class.

Interestingly, the student was caught because the article he submitted was so logically coherent and well-structured that the instructor only learned the truth after questioning him.

##△ Northern Michigan University

It needs to be made clear that the above packet capture was just accidental , and more often than not, the generated content has made it difficult for scientists to distinguish between authenticity and fiction.

A previous article in Nature pointed out that after the release of ChatGPT, a research team at Northwestern University in Illinois was exploring whether scientists could recognize the abstracts of medical papers generated using this AI tool.

The results found that all AI-generated summaries passed the plagiarism detector, and during manual review, 32% of the AI-generated summaries were considered to be completed by real humans.

Identification tools are already being developed

The nature’s failure shows that they attach great importance to the problems caused by ChatGPT.

In the latest content, Nature stated that the academic and research circles are worried that students and researchers may use content generated by large-scale language models as text written by themselves. In addition to taking risks, the above process also Can produce unreliable research conclusions.

Especially for ChatGPT, the Nature team and many publishers and platforms believe that this tool cannot be responsible for the integrity and content of scientific papers.

One side evidence is that the technology media CNET has exposed in the past two days that 41 of the 77 pieces of content written by the platform's AI have errors. At present, the platform has corrected them and stated that it will suspend the production of content in this way. .

It is precisely based on concerns about the above situation that Nature has issued relevant regulations.

As for how to distinguish AI-generated content?

Nature said that currently the raw output of ChatGPT can be discovered through careful inspection, especially when it comes to specific scientific work, the content may contain the simplest errors and general boring wording. In addition, Nature also said that they are still developing related recognition technology.

It is worth mentioning that relevant identification tools have also been developed.

For example, OpenAI’s own GPT-2 Output Detector can more accurately identify AI-generated text by inputting more than 50 characters (tokens).

Another example is Edward Tian, ​​a Princeton student, who also made a related tool GPTZero.

However, not all university professors and teachers have a negative attitude towards AI generation tools.

For example, Ethan Mollick, a professor at the Wharton School of the University of Pennsylvania, went the other way and required students to use ChatGPT to complete class assignments, believing that this was embracing emerging technology tools.

A worker at medRxiv, a paper publishing platform, also said that ChatGPT is not a new problem.

He believes that before, researchers have tried to sneak in the names of pets and fictional characters. Therefore, they believe that the core of the

problem is the need to continue to strengthen inspections.

Finally, have you used ChatGPT in your scientific research work?

The above is the detailed content of Nature's new rules: You can use ChatGPT to write papers, but not to be listed as an author. For more information, please follow other related articles on the PHP Chinese website!